Skip to main content

Data Cloud Integration

Execute queries and monitor AI data directly from GenAI Explorer with comprehensive Data Cloud integration for analytics and insights.

Data Cloud Integration

The Problem​

Analyzing AI performance requires querying Data Cloud to understand conversations, token usage, errors, and agent behavior.

Teams need to:

  • 📊 Monitor Performance: Track agent success rates, response times, and user satisfaction
  • 💰 Analyze Costs: Understand token consumption patterns and optimize spending
  • 🐛 Debug Issues: Investigate errors and failed conversations
  • 📈 Identify Trends: Spot patterns in user queries and agent responses
  • 🔍 Audit Behavior: Review what agents are doing and why
  • ⚡ Real-Time Insights: Get immediate visibility into AI operations

In short: You need quick access to AI data for analysis and optimization without writing complex SQL queries.

How GenAI Explorer Solves This​

GenAI Explorer provides instant data access with:

✅ Pre-Built Query Library: 8+ ready-to-use analytics queries

  • Complete reasoning traces
  • Token usage by model
  • Conversation threading
  • Feature usage statistics
  • Safety and compliance metrics
  • Error analysis
  • Performance trends

✅ Interactive Query Builder: Build custom queries with auto-complete

  • Object and field suggestions
  • Syntax validation
  • Query optimization tips

✅ Live Data Execution: Run queries directly against your org

  • Real-time results
  • No external tools needed
  • Secure OAuth authentication

✅ Multiple Export Formats: Get data where you need it

  • CSV for Excel
  • JSON for APIs
  • Text for reports
  • Copy to clipboard

✅ Query Lab in Atlas: Specialized queries for reasoning engine analysis

  • Stage-by-stage performance
  • ReAct loop iterations
  • RAG retrieval effectiveness

Impact: Analyze AI performance in minutes instead of days, export data instantly, and make data-driven optimization decisions without SQL expertise.

Overview​

GenAI Explorer provides deep integration with Salesforce Data Cloud, allowing you to query, analyze, and monitor all aspects of your AI implementation. Access real-time data about requests, responses, token usage, conversations, and performance metrics.

AI Data Objects​

GenAI Gateway Request​

Purpose: Incoming AI requests from users and agents.

Key Fields:

  • Id - Unique request identifier
  • InputText__c - User's input/query
  • SessionId__c - Conversation session ID
  • UserId__c - User who made the request
  • AgentId__c - AI Agent being queried
  • FeatureType__c - Type of AI feature used
  • ActionInvoked__c - Actions requested
  • CreatedDate - Timestamp of request
  • ModelRequested__c - AI model requested
  • Temperature__c - Model temperature parameter
  • MaxTokens__c - Maximum tokens allowed

Common Queries:

-- Recent requests
SELECT Id, InputText__c, CreatedDate, UserId__c
FROM GenAIGatewayRequest__c
WHERE CreatedDate = TODAY
ORDER BY CreatedDate DESC

-- Requests by user
SELECT UserId__c, COUNT(Id) as RequestCount
FROM GenAIGatewayRequest__c
WHERE CreatedDate = LAST_N_DAYS:7
GROUP BY UserId__c

-- Requests by agent
SELECT AgentId__c, COUNT(Id) as RequestCount
FROM GenAIGatewayRequest__c
WHERE CreatedDate = LAST_N_DAYS:30
GROUP BY AgentId__c

GenAI Gateway Response​

Purpose: AI-generated responses to requests.

Key Fields:

  • Id - Unique response identifier
  • RequestId__c - Foreign key to Request
  • OutputText__c - AI-generated response
  • CitedSources__c - References used
  • ResponseTimeMs__c - Time to generate
  • QualityScore__c - Quality assessment
  • SafetyScore__c - Safety assessment
  • PIIDetected__c - PII detection flag
  • PIIMasked__c - PII masking applied
  • ToxicityScore__c - Toxicity level
  • BiasScore__c - Bias detection score
  • ErrorMessage__c - Error details if failed
  • CreatedDate - Response timestamp

Common Queries:

-- Responses with citations
SELECT r.InputText__c, resp.OutputText__c, resp.CitedSources__c
FROM GenAIGatewayRequest__c r
JOIN GenAIGatewayResponse__c resp ON r.Id = resp.RequestId__c
WHERE resp.CitedSources__c != null
ORDER BY r.CreatedDate DESC

-- Quality analysis
SELECT 
  AVG(QualityScore__c) as AvgQuality,
  AVG(ResponseTimeMs__c) as AvgResponseTime,
  COUNT(Id) as TotalResponses
FROM GenAIGatewayResponse__c
WHERE CreatedDate = LAST_N_DAYS:7

-- Safety monitoring
SELECT Id, RequestId__c, SafetyScore__c, ToxicityScore__c
FROM GenAIGatewayResponse__c
WHERE SafetyScore__c < 0.8
   OR ToxicityScore__c > 0.3
ORDER BY CreatedDate DESC

GenAI Gateway Generation​

Purpose: Token-level details about AI generation.

Key Fields:

  • Id - Unique generation identifier
  • ResponseId__c - Foreign key to Response
  • Model__c - Actual model used
  • InputTokens__c - Input token count
  • OutputTokens__c - Output token count
  • TotalTokens__c - Total tokens used
  • CostEstimate__c - Estimated cost
  • ProcessingTimeMs__c - Processing time per stage
  • ModelParameters__c - Model settings used
  • CreatedDate - Generation timestamp

Common Queries:

-- Token usage by model
SELECT 
  Model__c,
  SUM(InputTokens__c) as TotalInput,
  SUM(OutputTokens__c) as TotalOutput,
  SUM(TotalTokens__c) as TotalTokens,
  AVG(TotalTokens__c) as AvgPerRequest,
  SUM(CostEstimate__c) as TotalCost
FROM GenAIGatewayGeneration__c
WHERE CreatedDate = LAST_N_DAYS:30
GROUP BY Model__c

-- Expensive requests
SELECT 
  g.ResponseId__c,
  g.TotalTokens__c,
  g.CostEstimate__c,
  r.InputText__c
FROM GenAIGatewayGeneration__c g
JOIN GenAIGatewayResponse__c resp ON g.ResponseId__c = resp.Id
JOIN GenAIGatewayRequest__c r ON resp.RequestId__c = r.Id
WHERE g.TotalTokens__c > 2000
ORDER BY g.TotalTokens__c DESC

Pre-Built Queries​

GenAI Explorer includes pre-built queries for common analysis tasks:

1. Complete Request-Response Trace​

Purpose: Follow the full lifecycle of an AI interaction.

Query:

SELECT 
  r.Id as RequestId,
  r.CreatedDate,
  r.InputText__c,
  r.SessionId__c,
  r.UserId__c,
  resp.OutputText__c,
  resp.ResponseTimeMs__c,
  resp.QualityScore__c,
  resp.CitedSources__c,
  gen.Model__c,
  gen.TotalTokens__c,
  gen.CostEstimate__c
FROM GenAIGatewayRequest__c r
JOIN GenAIGatewayResponse__c resp ON r.Id = resp.RequestId__c
JOIN GenAIGatewayGeneration__c gen ON resp.Id = gen.ResponseId__c
WHERE r.CreatedDate = TODAY
ORDER BY r.CreatedDate DESC
LIMIT 100

Use Case: Debug specific interactions, understand what happened end-to-end.

2. Token Usage Analysis​

Purpose: Analyze costs and optimize model selection.

Query:

SELECT 
  gen.Model__c,
  COUNT(DISTINCT r.SessionId__c) as SessionCount,
  COUNT(gen.Id) as RequestCount,
  SUM(gen.InputTokens__c) as TotalInputTokens,
  SUM(gen.OutputTokens__c) as TotalOutputTokens,
  SUM(gen.TotalTokens__c) as TotalTokens,
  AVG(gen.TotalTokens__c) as AvgTokensPerRequest,
  SUM(gen.CostEstimate__c) as TotalCost,
  AVG(gen.CostEstimate__c) as AvgCostPerRequest
FROM GenAIGatewayGeneration__c gen
JOIN GenAIGatewayResponse__c resp ON gen.ResponseId__c = resp.Id
JOIN GenAIGatewayRequest__c r ON resp.RequestId__c = r.Id
WHERE gen.CreatedDate = LAST_N_DAYS:30
GROUP BY gen.Model__c
ORDER BY TotalCost DESC

Use Case: Identify cost optimization opportunities.

3. Conversation Threading​

Purpose: Analyze multi-turn conversations.

Query:

SELECT 
  r.SessionId__c,
  COUNT(r.Id) as MessageCount,
  MIN(r.CreatedDate) as SessionStart,
  MAX(r.CreatedDate) as SessionEnd,
  TIMEDIFF(MAX(r.CreatedDate), MIN(r.CreatedDate)) as SessionDuration,
  AVG(gen.TotalTokens__c) as AvgTokensPerMessage,
  SUM(gen.TotalTokens__c) as TotalSessionTokens,
  AVG(resp.ResponseTimeMs__c) as AvgResponseTime
FROM GenAIGatewayRequest__c r
JOIN GenAIGatewayResponse__c resp ON r.Id = resp.RequestId__c
JOIN GenAIGatewayGeneration__c gen ON resp.Id = gen.ResponseId__c
WHERE r.CreatedDate = LAST_N_DAYS:7
  AND r.SessionId__c != null
GROUP BY r.SessionId__c
HAVING MessageCount > 1
ORDER BY MessageCount DESC, SessionStart DESC

Use Case: Understand conversation patterns and session behavior.

4. Performance Monitoring​

Purpose: Track response times and identify bottlenecks.

Query:

SELECT 
  r.AgentId__c,
  COUNT(resp.Id) as TotalResponses,
  AVG(resp.ResponseTimeMs__c) as AvgResponseTime,
  MIN(resp.ResponseTimeMs__c) as MinResponseTime,
  MAX(resp.ResponseTimeMs__c) as MaxResponseTime,
  PERCENTILE(resp.ResponseTimeMs__c, 0.5) as MedianResponseTime,
  PERCENTILE(resp.ResponseTimeMs__c, 0.95) as P95ResponseTime,
  AVG(resp.QualityScore__c) as AvgQualityScore
FROM GenAIGatewayRequest__c r
JOIN GenAIGatewayResponse__c resp ON r.Id = resp.RequestId__c
WHERE r.CreatedDate = LAST_N_DAYS:7
GROUP BY r.AgentId__c
ORDER BY AvgResponseTime DESC

Use Case: Identify slow agents and optimize performance.

5. Feature Usage Analysis​

Purpose: Track which AI features are most used.

Query:

SELECT 
  r.FeatureType__c,
  r.ActionInvoked__c,
  COUNT(r.Id) as UsageCount,
  COUNT(DISTINCT r.UserId__c) as UniqueUsers,
  AVG(resp.ResponseTimeMs__c) as AvgResponseTime,
  AVG(resp.QualityScore__c) as AvgQuality,
  COUNT(CASE WHEN resp.ErrorMessage__c != null THEN 1 END) as ErrorCount,
  (COUNT(CASE WHEN resp.ErrorMessage__c != null THEN 1 END) * 100.0 / COUNT(r.Id)) as ErrorRate
FROM GenAIGatewayRequest__c r
JOIN GenAIGatewayResponse__c resp ON r.Id = resp.RequestId__c
WHERE r.CreatedDate = LAST_N_DAYS:30
GROUP BY r.FeatureType__c, r.ActionInvoked__c
ORDER BY UsageCount DESC

Use Case: Understand feature adoption and prioritize improvements.

6. Safety & Compliance Monitoring​

Purpose: Track PII detection, masking, and safety scores.

Query:

SELECT 
  r.CreatedDate,
  r.UserId__c,
  r.AgentId__c,
  resp.SafetyScore__c,
  resp.PIIDetected__c,
  resp.PIIMasked__c,
  resp.ToxicityScore__c,
  resp.BiasScore__c,
  CASE 
    WHEN resp.SafetyScore__c < 0.7 THEN 'High Risk'
    WHEN resp.SafetyScore__c < 0.85 THEN 'Medium Risk'
    ELSE 'Low Risk'
  END as RiskLevel
FROM GenAIGatewayRequest__c r
JOIN GenAIGatewayResponse__c resp ON r.Id = resp.RequestId__c
WHERE r.CreatedDate = LAST_N_DAYS:30
  AND (resp.SafetyScore__c < 0.8 
       OR resp.PIIDetected__c = true
       OR resp.ToxicityScore__c > 0.3
       OR resp.BiasScore__c > 0.3)
ORDER BY resp.SafetyScore__c ASC, r.CreatedDate DESC

Use Case: Compliance auditing and safety monitoring.

7. User Engagement Analysis​

Purpose: Track user adoption and engagement.

Query:

SELECT 
  r.UserId__c,
  COUNT(DISTINCT DATE(r.CreatedDate)) as ActiveDays,
  COUNT(r.Id) as TotalRequests,
  COUNT(DISTINCT r.SessionId__c) as TotalSessions,
  AVG(resp.QualityScore__c) as AvgQualityScore,
  SUM(gen.TotalTokens__c) as TotalTokensUsed,
  MIN(r.CreatedDate) as FirstRequest,
  MAX(r.CreatedDate) as LastRequest
FROM GenAIGatewayRequest__c r
JOIN GenAIGatewayResponse__c resp ON r.Id = resp.RequestId__c
JOIN GenAIGatewayGeneration__c gen ON resp.Id = gen.ResponseId__c
WHERE r.CreatedDate = LAST_N_DAYS:30
GROUP BY r.UserId__c
ORDER BY TotalRequests DESC

Use Case: Understand user adoption and identify power users.

8. Error Analysis​

Purpose: Identify and track errors.

Query:

SELECT 
  DATE(r.CreatedDate) as ErrorDate,
  r.AgentId__c,
  resp.ErrorMessage__c,
  COUNT(resp.Id) as ErrorCount,
  COUNT(DISTINCT r.UserId__c) as AffectedUsers
FROM GenAIGatewayRequest__c r
JOIN GenAIGatewayResponse__c resp ON r.Id = resp.RequestId__c
WHERE r.CreatedDate = LAST_N_DAYS:7
  AND resp.ErrorMessage__c != null
GROUP BY DATE(r.CreatedDate), r.AgentId__c, resp.ErrorMessage__c
ORDER BY ErrorDate DESC, ErrorCount DESC

Use Case: Track and prioritize bug fixes.

SOQL Query Builder​

Features​

Interactive Query Editor:

  • Syntax highlighting
  • Auto-completion
  • Field suggestions
  • Error detection

Result Display:

  • Sortable columns
  • Filterable data
  • Pagination
  • Export to CSV/Excel

Query History:

  • Save favorite queries
  • Browse recent queries
  • Share queries with team

Using the Query Builder​

  1. Select Object

    • Choose from AI data objects
    • View object schema
    • See field descriptions

  2. Build Query

    • Select fields to retrieve
    • Add WHERE conditions
    • Set ORDER BY
    • Configure LIMIT

  3. Execute

    • Click Run Query
    • View results
    • Analyze data

  4. Export

    • Download as CSV
    • Copy to clipboard
    • Save to board

Visualizations​

Create Charts from Query Results​

Supported Chart Types:

  • Line charts (time series)
  • Bar charts (comparisons)
  • Pie charts (distributions)
  • Scatter plots (correlations)
  • Heatmaps (patterns)

Example: Token Usage Over Time

SELECT 
  DATE(CreatedDate) as Date,
  SUM(TotalTokens__c) as TotalTokens
FROM GenAIGatewayGeneration__c
WHERE CreatedDate = LAST_N_DAYS:30
GROUP BY DATE(CreatedDate)
ORDER BY Date

Visualization: Line chart showing daily token usage trends.

Dashboard Creation​

Combine multiple queries into a dashboard:

Components:

  • Metric cards (KPIs)
  • Charts and graphs
  • Data tables
  • Live updates

Example Dashboard:

  • Total requests today
  • Average response time
  • Token usage by model
  • Top 10 users
  • Recent errors